Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

raymondu999

macrumors 65816
Original poster
Feb 11, 2008
1,009
1
Hey guys. Right now, everybody is probably mostly thinking about the upcoming Montevina+Penryn update (if any) for Macs. Now, there's also, as many of you should know, Core i7 next year, eliminating the FSB. Now, this effectively moves the bottleneck from FSB speed to Processor clock speed, correct?

What I wanted to ask is, actually, nowadays, is there actually still an FSB bottleneck at 1066MHz? I mean, if there isn't actually any real constraint at 1066MHz, wouldn't that mean the QuickPath Interconnect architecture would bring, at this point, close to no performance gains? I mean, sure, there's bound to be some power consumption decrease, and there's also bound to be some performance gains from the returned multithreading feature... but... really, other than that, does the QPI (QuickPath Interconnect) really bring more performance benefits at this age, when compared to 1066MHz?

Just wanted some opinions... Of course, naturally, informed answers are also welcome:D
 
Look up the benchmarks of the early processor prototypes on Google and decide for yourself how big of an improvement it is. :)
 
I would wager there aren't many iMac, Macbook Pro, Macbook or Mini users that are suffering from the limits of the FSB. QPI is more a solution for servers and workstations.
 
Hey guys. Right now, everybody is probably mostly thinking about the upcoming Montevina+Penryn update (if any) for Macs. Now, there's also, as many of you should know, Core i7 next year, eliminating the FSB. Now, this effectively moves the bottleneck from FSB speed to Processor clock speed, correct?

What I wanted to ask is, actually, nowadays, is there actually still an FSB bottleneck at 1066MHz? I mean, if there isn't actually any real constraint at 1066MHz, wouldn't that mean the QuickPath Interconnect architecture would bring, at this point, close to no performance gains? I mean, sure, there's bound to be some power consumption decrease, and there's also bound to be some performance gains from the returned multithreading feature... but... really, other than that, does the QPI (QuickPath Interconnect) really bring more performance benefits at this age, when compared to 1066MHz?

The FSB clock speed is just a clock speed, in other words, a meaningless marketing number. Just like the revs on your car. Take a Porsche in first gear, rev it to the limit, and I will easily overtake it in fifth gear at 2000 revs. The bottleneck is the access speed of the memory connected, how many bits the memory can read or write in parallel, and how long the FSB takes to figure out what to do. With the direct connection without FSB, the part "how long the FSB takes to figure out what to do" is gone, that is a substantial improvement.

Next, the Nehalem chip can have _three_ memory connections. That means the same chip can read or write from up to three different sets of memory simultaneously. It will depend on how clever the memory is divided up between different memory chips, but this can increase throughput up to a factor three. Now if you had a single core, that might not make much difference. But with a quad core chip you can have four processes demanding memory, so being able to read from three sets of memory chips at the same time should make an awful lot of difference.

Quickpath enters the game if you have more than one physical chip. On all current Macs with two physical chips, these chips may have to fight about how can access memory at one time. This slows things down; the fact alone that one chip has to check that the other one isn't reading at the same time makes things slow. With Quickpath, there is a direct connection between the chips. Instead of fighting about which chip can read the memory, each chip has some memory connected exclusively to itself, and it sends requests for data connected to the other chips over Quickpath. No fighting means improved speed for systems with two or four physical chips; that would be eight or sixteen core systems.
 
I realise that yes, in reality Nehalem WILL bring about performance gains, and benchmarks will show quite a leap. But what I'm asking is more of, CAN WE FEEL these performance gains?
 
Next, the Nehalem chip can have _three_ memory connections.
Just to clarify things, the regular desktop and mobile variants will still have two memory channels.

The latest signs points to those variants of Nehalem coming in Q4 2009 for quad-core and January 2010 for dual-core. :mad::( It's also hinted that Sandy Bridge may be delayed for "sufficient Nehalem lifecycle." :rolleyes: The only good news is that mobile Nehalem has been rumored to be on 32 nm rather than 45 nm.

lynnfield.png
 
For word editing and playing minesweeper...no, you won't feel the difference (compared to a computer of today)

for gaming, servers, yes you will feel the difference
 
I realise that yes, in reality Nehalem WILL bring about performance gains, and benchmarks will show quite a leap. But what I'm asking is more of, CAN WE FEEL these performance gains?

Every time we have performance gains, they enable software to do things that couldn't be done before. Look at the iPhone software. What this software does is only possible because the iPhone has two processors, each more powerful than the first generation of iMacs, with a lot more RAM and more permanent storage. Plus a graphics accelerator, which that first iMac didn't have.

There will be applications that make an eight core Nehalem machine struggle; in five years these applications will be mainstream.
 
Whoa whoa whoa, I was under the impression that mobile nehalem would be out around next summer. Has intel changed this up?
 
Nehalem is now targeted for Q3 2009. Actually, the June 2007 (Santa Rosa) MacBook Pros already supported 8GB, hardware-wise. It's just that: 1) there are so little (though they ARE available) 4GB SO-DIMMs around and 2) Mac OS X, in the MacBook Pro, restricts the accepted RAM to 4GB, so that you cannot take in 8GB.

Hence, if suddenly Apple releases a software update enabling 8GB RAM, then we can, in theory, support 8GB on Santa Rosa, and Penryn. So the hardware is there, but it's just the software side...
 
Errm... I'm lost. Care to explain?
In June the rumors were flying that all mainstream (Lynnfield, Clarksfield, Havendale, Auburndale) Nehalem CPUs were delayed from Q1/Q2 2009 to Q3 2009.

The link and its image I posted here is from this month and it shows a further delay of mainstream Nehalem by 1/2 months.
 
So the next macbook/pro update is currently rumoured to be Penryn to Montevina and a case resdesign in October 2008? :confused:
 
Hope we'll have an idea of when they're coming out with a better release date come the next IDF...
 
Ic... @lasuther, if you noticed, I was referring to the launch date of the chips+Chipset themselves, not the new MBPs:p
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.